Fizzer you done good!

You have quite a community of programmers here who want to help make Warlight better and you finally made something open-source to improve a feature for everyone.

Personally I like doing weighted randoms

*cries for MoD*

But I am gonna have a look at this, maybe change it's picking style, as I think that is where I should be the strongest.

Edited 2/25/2016 18:21:08

I can see problem with fixed starting locations (like full distribution or non-random warlords)

Certainly -- that bot was written to compete on TheAIGames, so it was never taught to pick in those kinds of situations. It's totally understandable, and expected, that it will fail in many situations.

In my opinion starting point should be ["bonus value"/"no. of territories"] variable, and then modified by other factors.

I believe it currently uses "bonus value" / "number of neutral armies in the bonus", and then modifies that by other factors. This helps it account for wastelands.

I believe it currently uses "bonus value" / "number of neutral armies in the bonus", and then modifies that by other factors. This helps it account for wastelands.

That's not ideal for when you're using zero neutral armies. Surely "bonus value" / ("number of neutral armies in the bonus"+1) would make more sense.

Edited 2/25/2016 18:41:19

Ok, so how strong exactly were the AI's that won previous competitions? Were they even close to competing vs humans?

I just ran the .exe file. My AI played one turn however then an exception occurred:

(The first german line means that the input string sequence isn't a valid DateTime)
-----
System.FormatException: Die Zeichenfolge wurde nicht als gültiges DateTime erkan nt.
bei System.DateTimeParse.Parse(String s, DateTimeFormatInfo dtfi, DateTimeSty les styles)
bei System.DateTime.Parse(String s)
bei WarLight.AI.Communication.ReadGameTurn(JToken jToken) in D:\Stuff\WarLigh t.AI\Common\Comm\Communication.cs:Zeile 183.
bei WarLight.AI.Communication.ReadLatestInfo(JToken gameInfo) in D:\Stuff\War Light.AI\Common\Comm\Communication.cs:Zeile 46.
bei WarLight.AI.Communication.ReadGameObject(JToken response, GameIDType game ID) in D:\Stuff\WarLight.AI\Common\Comm\Communication.cs:Zeile 84.
bei WarLight.AI.HumanGameAPI.GetGameInfo(GameIDType gameID, Nullable`1 turnNu mber) in D:\Stuff\WarLight.AI\Common\Comm\HumanGameAPI.cs:Zeile 76.
bei WarLight.AI.PlayHuman.PlayLoop(String botName, GameIDType gameID, PlayerI DType playerID) in D:\Stuff\WarLight.AI\Common\PlayHuman.cs:Zeile 69.
bei WarLight.AI.PlayHuman.Create(String botName, String opponent, String game Name) in D:\Stuff\WarLight.AI\Common\PlayHuman.cs:Zeile 32.
bei WarLight.AI.EntryPoint.Go(String[] args) in D:\Stuff\WarLight.AI\Common\E ntryPoint.cs:Zeile 76.
bei WarLight.AI.EntryPoint.Main(String[] args) in D:\Stuff\WarLight.AI\Common \EntryPoint.cs:Zeile 21.

cle, then change the bot to how you would do it ;)

swe, On the specific template they were built for they did pretty well, most people agree that wafflewaltz came at around a 1700. The issue here is we now need to change the code to suit any template, which is a lot more tricky, and also to deal with FFAs, which (assuming we want the AI to play smart) is a lot different.

I just ran the .exe file. My AI played one turn however then an exception occurred:

I ran into the same exception. It seems to fail on German, Russian and any other non-US date format. I managed to work around the exception though. Tell the program that the system locale is US English.

$ LANG=C mono WarLight.AI.exe

Edited 2/25/2016 19:56:59

could someone explain me how to make an Al?

could someone explain me how to make an Al?

https://www.warlight.net/blog/index.php/2016/02/the-warlight-ai-goes-open-source/

could someone explain me how to make an Al?



https://www.warlight.net/blog/index.php/2016/02/the-warlight-ai-goes-open-source/

I just ran the .exe file. My AI played one turn however then an exception occurred:

(The first german line means that the input string sequence isn't a valid DateTime)

Sorry about that. The DateTimes are transmitted in US format, so it broke on a non-US system.

I just made a change to the code to specify that the DateTime is in en-US format. I don't have a non-US OS to test, but I believe this will fix it. Are you able to get the latest code from github, build it, and see if the problem is fixed?

Are you able to get the latest code from github, build it, and see if the problem is fixed?

I have tested the patch in Russian environment. Runs good.

could you make the AI surrender when there is no way that it can win?

could you make the AI surrender when there is no way that it can win?

It will follow the surrender settings after it is developed, tested and deployed in production. It should not surrender otherwise. AI is never strong in computing if it can not win.

So, who is planning on working on this AI? I wrote my bot mainly between University and my first job, nowadays I don't have time for "hardcore coding" under the week. However I'd still like to participate.

I haven't looked into the details yet. Do the bots have fog? Apart from the Minimax picking (which Fizzer removed) calculating what is behind the fog was the only time consuming operation (and with reinforcement cards involved it would get quite more difficult). Apart from that, since Wunderwaffe doesn't look into further configurations, it should be capable to play on every map size very fast.

Also Wunderwaffe shouldn't have a concept of teammates or different opponents in ffa, obviously Fizzer had to adapt some stuff there (Teammates = neutrals?).

There will be a lot of work to create a good AI that can be hard coded into Warlight as the main one. The think the idea of a "fun" AI is that it won't be too strong.

I mean, idealy we can code into it responses depending on how tough the human it's playing against is (detecting behaviour)

The fact it can't learn is a possible problem in terms of not being able to adapt to players specific strengths and weaknesses and "go easy" on players it beats.

But I think it we all work together to some degree we can develop quite a robust little package that can hopefully deal with lots of various settings too.

I'm particularly interested in developing it's ability to use cards. I'll have to really start looking at this properly in the next few days, and seeing what we can do :)

Semi, by the sounds of things, you will always be able to run your own or (I presume) other people's AI scripts, but Fizzer is talking about specifically an AI to add to replace the current one. There's nothing to stop more being added in the future, but this is a hige step, because the current AI is way under-par and is holding back a lot of potential in the game.

I can help with testing. Although, I don't know any specific language, I have general idea about programming, algorithms etc, so I can translate problem I see to form understandable for programmer.

EDIT: actually now when I think of it, I could even cooperate with a programmer. I can help with creating all the algorithms.

Edited 2/26/2016 10:35:26

The fact it can't learn is a possible problem in terms of not being able to adapt to players specific strengths and weaknesses and "go easy" on players it beats.

Actually, I think it's impossible to crate AI strong enough to be really challenging even for a beginner. After you grasp basic rules, it's not hard to outsmart AI. And in my opinion AI, that actually require you to think is much more fun to play, than the one, that can beat you only with huge initial advantage (eg. current Insane SP challenge).

Also I would like to extend an idea, that was mentioned before. It might be interesting to create several AI's each for a specific settings, and then combine them together. So for example, there would be separate AI used for FFA's, 1v1 games, team games, maybe even different ones for game witch cards or full distribution/warlord distribution.

in this way, each person could write part of the code and create together AI that is versatile.

technically speaking you could write wrappers for other languages, but the example is aimed at c# development.